Performance of OLCI Sentinel-3A satellite in the Northeast Pacific coastal waters

Highlights

• C2RCC and POLYMER algorithms were evaluated for OLCI Sentinel-3 in a coastal area.

• TSM concentrations from POLYMER provided the lowest differences from in situ data.

• Chl-a estimates from POLYMER were satisfactory, but need specific data quality flags.

• CDOM was accurately estimated from ADG443 retrieved by C2RCC v2 (extended NN).

• OLCI data adequately reflect the expected seasonality at the Northeast Pacific coast.

Abstract

Coastal oceans play a pivotal role in fisheries production and global biogeochemical cycles, making large-scale monitoring an essential task. The advent of modern remote sensors, such as OLCI (Ocean Land Colour Instrument), on board the Sentinel-3A satellite, has made it possible to obtain oceanic biogeochemical products at higher spatial (300 m) and temporal (daily) resolutions than previously possible. However, validating the Sentinel-3A retrievals for coastal waters is an ongoing effort. Using a regional in situ dataset from British Columbia (BC) and Southeast Alaska (SEA) coastal waters, we evaluated the performance of OLCI Sentinel-3A in retrieving remote-sensing reflectance (Rrs) and biophysical products, including total suspended matter (TSM), chlorophyll-a concentration (Chl-a) and coloured dissolved organic matter (CDOM). The OLCI data were processed through a spectral optimization-based algorithm (POLYMER) and a neural net-based (NN) algorithm (C2RCC), including the original (C2RCC v1) and the alternative version (altNN or v2), in which the neural network was trained with extended ranges to cope with larger dynamic range for high backscatter waters. The processors' performance was evaluated through match-up analysis using data from southern BC, as well as expected ranges and seasonal trends for northern BC and SEA. Multimetric statistical analyses demonstrated that POLYMER provided the best overall performance for TSM and Chl-a retrievals, with the Chl-a product improved by the use of the so-called “Case-2” flag. Despite the relative outperformance of POLYMER-derived products, with low systematic biases, the relative percent differences are still high (80–100%) and should be acknowledged in future analyses when using these data. CDOM, only retrieved here from NN approaches, was better estimated using the alternative version of C2RCC. The best performing approaches were used to evaluate Level-3 composites for northern BC and SEA waters. The observed spatial and seasonal trends compared favourably with those reported in the literature, including highlighting more productive areas (e.g. west coast of Vancouver Island), and the important interactions between riverine systems and adjacent coastal waters (e.g., high TSM and CDOM loads near the Skeena and Nass rivers). This study demonstrated the benefits of OLCI Sentinel-3A to investigate complex coastal ecosystems and provides a robust evaluation of OLCI performance and a framework for future observational and process-oriented studies.

Introduction

Remotely sensed data of the oceans' surface have been increasingly used to monitor and understand biogeochemical processes that might influence or be associated with the changes in the Earth's climate (Behrenfeld et al., 2006; Behrenfeld et al., 2016; Sathyendranath et al., 2017; Dutkiewicz et al., 2019). Since the launch of the NASA Coastal Zone Colour Scanner (CZCS) on board the Nimbus-7 in 1978, ocean colour sensors have been used to provide biogeophysical and water quality parameters at higher spatial and temporal resolution compared to the field data (McClain, 2009; Groom et al., 2019). Satellite retrievals include chlorophyll-a concentration (a proxy for phytoplankton biomass; e.g., Mélin et al., 2011; Lavender et al., 2015), load and sources of dissolved organic material (e.g., Nelson and Siegel, 2013; Mannino et al., 2014), and levels of turbidity and sediment transport (e.g., Nechad et al., 2010; Dogliotti et al., 2015). However, to date, ocean colour remote sensing has most effectively been applied in off-shelf waters, while its application in coastal waters remains challenging, inhibiting the retrieval of accurate satellite-derived products (IOCCG, 2000; Park and Ruddick, 2005; Werdell et al., 2018).

Optically, coastal waters differ from the open ocean due to the presence of non-algal particles and dissolved material, in addition to the phytoplankton cells, which together interact with (i.e., absorb and scatter) the incident light in the ocean (Prieur and Sathyendranath, 1981). The concurrent presence of several optically active constituents modulates the reflectance signal leaving the surface water and detected by remote sensors, which is ultimately used to inversely derive products through bio-optical models (Mobley, 1994; Morel and Maritorena, 2001). Thus, in coastal waters, especially when turbid freshwater outflows are present, the development of coastal bio-optical models, as well as regional validation of existing models, is required and has indeed been performed by a number of research groups worldwide (e.g., Moore et al., 1999; Nechad et al., 2010; Giannini et al., 2013; Tilstone et al., 2017; Carswell et al., 2017). Furthermore, quantifying the parameters required for the removal of the atmospheric signals from the top of the atmosphere radiance data measured by satellite-borne sensors is more complex for turbid coastal waters than for oceanic waters (Siegel et al., 2000), affecting the performance of atmospheric correction models. In clear waters, the assumption that seawater is totally absorbent in the near-infrared (NIR) spectral bands is accepted, and the aerosol optical properties are then modelled for the visible bands (Gordon and Wang, 1994). However, this assumption is invalid in turbid or non-phytoplankton-dominated waters (Ruddick et al., 2000; Goyens et al., 2013), and the aerosol content in the atmosphere is rather inferred from signals in the NIR or the short-wavelength infrared region (SWIR) or a combination of SWIR/NIR (Wang and Shi, 2005; Ruddick et al., 2000; Carswell et al., 2017). Alternatively, atmospheric correction algorithms have been developed based on iterative coupled ocean-atmosphere models, such as POLYMER (POLYnomial based algorithm applied to MERIS), based on polynomial spectral matching technique (Steinmetz et al., 2011) and C2RCC (Case-2 Regional Coast Colour), based on artificial neural network simulations (Doerffer and Schiller, 2007; Brockmann et al., 2016). Further details about the ocean-atmosphere coupled models are provided in Section 2.3.

Beyond their optical complexity, coastal environments can have a high degree of geomorphological and oceanographic complexity due to small-scale dynamic processes such as tidal cycles, riverine discharge and frontal systems, and capturing these dynamics requires satellite data at fine spatial (few meters) and temporal (hours to daily) resolutions (IOCCG, 2000). Therefore, in addition to the development of bio-optical and atmospheric correction models tuned to coastal conditions, instrument-related advances have tended towards sensors capable of dealing with optically and oceanographically complex waters. One such advancement is the OLCI sensor, part of the Sentinel-3 mission jointly operated by ESA and EUMETSAT, which has been operational since April 2016. The Sentinel-3 OLCI delivers data from the surface ocean at 300 m resolution and was designed to continue the Envisat MERIS mission. The OLCI sensor has an improved signal-to-noise ratio and off-nadir swath centered to minimize sun glint (Donlon et al., 2012). Sentinel-3 OLCI is therefore particularly well suited to studies of complex nearshore waters, with potential applications in both research and management.

A critical first step before the broader application of OLCI Sentinel-3 is the validation of target products. Indeed, since its launch, OLCI's performance and associated uncertainties have been under continuous evaluation, especially for coastal waters (e.g., Hieronymi et al., 2017; Zibordi et al., 2018; Mograne et al., 2019; Gossn et al., 2019). However, due to its relatively recent launch, there is still extensive discussion about the deviations found by the different processing approaches applied to Sentinel-3A OLCI imagery, and consequently derived products. The aim of this study was to evaluate the performance of this sensor to retrieve biogeophysical products (chlorophyll-a concentration, total suspended material, and coloured dissolved organic matter) in the optically complex coastal waters of British Columbia (BC, Canada) and Southeast Alaska (SEA, US). The satellite-derived products were generated through different atmospheric correction approaches, and their performances were statistically validated against in situ data. In addition, in order to evaluate the application of the methods for large-scale imagery of the Northeast Pacific coastal waters, the retrieved data were discussed in terms of expected seasonal and latitudinal dynamics for the study area.